ERROR: invalid page header in block 27073 of relation base/21078/45300926

本文涉及的产品
云原生数据库 PolarDB MySQL 版,Serverless 5000PCU 100GB
简介:
突然断网,检查后通知我们UPS断电,db所在主机重启 
1、连上后,发现pg主从不同步,主不向从传日志,从报错: 
FATAL: could not connect to the primary server: could not connect to server: 
发现从先启动成功,而主是后启动的,因此我们将从再次重启 
service postgresql restart 
开始正常传日志 

2、过了一会,研发反应部分表的使用出现问题,主再次不传输日志,且无sender进程。 
查看主库日志,发现报错: 
ERROR:  invalid page header in block 27073 of relation base/21078/45300926 
查看从库日志,发现类似报错 
ERROR:insert_t_black:invalid page head in block 27073 of relation base/21078/45300926 

21078是tm_samples库的id号 
通过查询: 
select * from pg_class where relfilenode='45300926';—t_black 
应该是停电的时候,正好在进行数据的写入等操作,造成了数据损坏。 

此时,我们进行了如下操作: 
set  zero_damaged_pages = on;   --当这个参数为on的时候,会忽略所有数据有损坏的页面 
vacuum full t_black; 
tm_samples=# select count(*) from t_black; 
WARNING: invalid page header in block 27069 of relation base/21078/45300926; zeroing out page 
WARNING: invalid page header in block 27070 of relation base/21078/45300926; zeroing out page 
WARNING: invalid page header in block 27071 of relation base/21078/45300926; zeroing out page 
WARNING: invalid page header in block 27072 of relation base/21078/45300926; zeroing out page 
WARNING: invalid page header in block 27073 of relation base/21078/45300926; zeroing out page 
count 
--------- 
3066302 
(1 row) 
reindex table t_black; 
再次select count(*) from t_black;  --正常 
再重启pg主从的服务,发现主从开始同步,后台没有再报刚刚那个错误 

3、研发测试验证,发现插入部分hash值的时候,依然会刚刚那个错误 
insert into t_black  (sample_hash, sample_crc32, sample_sha1, virusname, avl_m, software_english,file_size, conditions_type, 
sample_type, description,record_time, getname_time, developers, last_record_time, last_changename_time, source_id, ss_id, cps,level,keyhash, 
sensitive_strings, behavior_info, relation_info,version,ratio,"AVL_Embed", "AVL_Adware", program_name, versionname, record_time_int ) SELECT sample_hash, sample_crc32, sample_sha1, virusname, avl_m, software_english,file_size, conditions_type, 
sample_type, description,record_time, getname_time, developers, last_record_time, last_changename_time, source_id, ss_id, cps,level,keyhash, 
sensitive_strings, behavior_info, relation_info,version,ratio,"AVL_Embed", "AVL_Adware", program_name, versionname, record_time_int 
  FROM t_white where sample_hash ='1E0CB07CDC71B2F994F5D3EB51050E3A'; 

程序报错:DatabaseError: invalid page header in block 27073 of relation base/21078/45300926 
只插入部分字段就没问题,但完整插入就不行。 

这时,我们想到可能是文件系统损坏了,当有数据往刚刚那个坏块里面写的时候,就会触发报错。 
同时,我们查到一共有4张表出现了坏块,分别是: 
t_black 
t_white 
t_batch_sample 
t_derivative 
我们决定先对所有的表做一次vacuum,然后进行磁盘的检查------后面的操作都是娟姐做的,待她回来后补充,但她反应检查后依然不行,最后的方法是磁盘格式化,然后用从库的备份进行了覆盖恢复。 

网页上有一些介绍的方法: 
1)根据错误提示 ERROR: invalid page header in block 1 of relation base/34780/34781 我们可以找到相应的文件, 文件的路径为: 数据目录/base/34780/34781,只要用工具手动把上面提示的坏块清除即可。 
在Linux下面可以用dd工具把相应的页面清除: 
$dd if=/dev/zero of=/home/postgres/data/base/34780/42995 bs=8192 seek=1 count=1 conv=notrunc 

2)数据库暂时恢复使用了,但这是临时的,如果是磁盘文件系统故障,不久还是会重现这个问题,临了再修复了一下 
1.reboot进入单用户模式 
2.umount出现数据库异常的磁盘 
3.fsck -v -t -p /dev/sda1 
4.reboot 

因为把有坏块的主库留了个备份,想用这个备份做一些实验: 

a.尝试找到更为直接修复的方法 
b.换到其他硬盘好的地方去,看它到底是db本身的文件坏了,还是啥原因。 

======================实验======================= 
将/data和pg_xlog压缩,传到已经安装好pg的12.250上面。替换到原来的data目录,目录保留 
-rw-r--r-- 1 root root 39966776421 Apr  1 17:57 bk_data20150331.tar.gz 
-rw-r--r-- 1 root root  5475189513 Apr  1 18:18 bk_pg_xlog20150331.tar.gz 

[root@pgtina 9.1]# ll 
total 39030084 
drwxr-xr-x.  2 postgres postgres        4096 Apr  1 14:34 bin 
drwx------. 13 postgres postgres        4096 Apr  1 16:22 bk_data   ---原来的data目录 
-rw-r--r--.  1 root     root     39966776421 Apr  1 18:59 bk_data20150331.tar.gz 
drwx------. 14 postgres postgres        4096 Apr  2 14:36 data      ---解压后的data目录 
drwxr-xr-x.  6 postgres postgres        4096 Apr  1 14:34 include 
drwxr-xr-x.  3 postgres postgres        4096 Apr  1 14:34 lib 
drwxr-xr-x.  6 postgres postgres        4096 Apr  1 14:34 share 
[root@pgtina 9.1]# pwd 
/home/pgsql/9.1   ---跟原来的库安装路径保持一致。 

[postgres@pgtina ~]$ pg_ctl -D /home/pgsql/9.1/data start  --直接启动报错:参数设置太大,需要调小 
server starting 
[postgres@pgtina ~]$ 2015-04-02 14:26:31 CST--- :FATAL:  could not create shared memory segment: Cannot allocate memory 
2015-04-02 14:26:31 CST--- icon_biggrin.gif ETAIL:  Failed system call was shmget(key=5432001, size=8836481024, 03600). 
2015-04-02 14:26:31 CST--- :HINT:  This error usually means that PostgreSQL's request for a shared memory segment exceeded available memory or swap space, or exceeded your kernel's SHMALL parameter.  You can either reduce the request size or reconfigure the kernel with larger SHMALL.  To reduce the request size (currently 8836481024 bytes), reduce PostgreSQL's shared memory usage, perhaps by reducing shared_buffers or max_connections. 
The PostgreSQL documentation con 

[postgres@pgtina data]$ vi postgresql.conf 
[postgres@pgtina data]$ pg_ctl -D /home/pgsql/9.1/data start 
server starting 
[postgres@pgtina data]$ 2015-04-02 06:31:48 GMT--- :FATAL:  WAL archival (archive_mode=on) requires wal_level "archive" or "hot_standby"  --开了归档模式,改成off 
^C 
[postgres@pgtina data]$ vi postgresql.conf 
[postgres@pgtina data]$ pg_ctl -D /home/pgsql/9.1/data start 
server starting 
[postgres@pgtina data]$ 2015-04-02 06:32:12 GMT--- :FATAL:  WAL streaming (max_wal_senders > 0) requires wal_level "archive" or "hot_standby" ---改为0,不需要传输日志。 


[postgres@pgtina pg_log]$ more postgresql-2015-04-02_143234.log      ---一定要有流日志才能进行启动,因此再讲pg_xlog也解压 
2015-04-02 14:32:34 CST--- :LOG:  database system was shut down at 2015-03-30 22:26:53 CST 
2015-04-02 14:32:34 CST--- :FATAL:  required WAL directory "pg_xlog" does not exist 
2015-04-02 14:32:34 CST--- :LOG:  startup process (PID 11993) exited with exit code 1 
2015-04-02 14:32:34 CST--- :LOG:  aborting startup due to startup process failure 
[postgres@pgtina pg_log]$ cd .. 


[postgres@pgtina pg_log]$ more postgresql-2015-04-02_143335.log 
2015-04-02 14:33:35 CST--- :LOG:  database system was shut down at 2015-03-30 22:26:53 CST 
2015-04-02 14:33:35 CST--- :LOG:  creating missing WAL directory "pg_xlog/archive_status" 
2015-04-02 14:33:35 CST--- :LOG:  could not open tablespace directory "pg_tblspc/30139248/PG_9.1_201105231": No such file or directory 
2015-04-02 14:33:35 CST--- :LOG:  could not open file "pg_xlog/0000000100000B6400000027" (log file 2916, segment 39): No such file or directory 
2015-04-02 14:33:35 CST--- :LOG:  invalid primary checkpoint record 
2015-04-02 14:33:35 CST--- :LOG:  could not open file "pg_xlog/0000000100000B6400000025" (log file 2916, segment 37): No such file or directory 
2015-04-02 14:33:35 CST--- :LOG:  invalid secondary checkpoint record 
2015-04-02 14:33:35 CST--- icon_razz.gif ANIC:  could not locate a valid checkpoint record 
2015-04-02 14:33:35 CST--- :LOG:  startup process (PID 12008) was terminated by signal 6: Aborted 
2015-04-02 14:33:35 CST--- :LOG:  aborting startup due to startup process failure 


再次启动,虽有一个报错,但可以正常连接打开数据库了 
[root@pgtina pg_log]# more postgresql-2015-04-02_151734.log 
2015-04-02 15:17:34 CST--- :LOG:  database system was shut down at 2015-03-30 22:26:53 CST 
2015-04-02 15:17:34 CST--- :LOG:  could not open tablespace directory "pg_tblspc/30139248/PG_9.1_201105231": No such file or direc 
tory 
2015-04-02 15:17:34 CST--- :LOG:  database system is ready to accept connections 
2015-04-02 15:17:34 CST--- :LOG:  autovacuum launcher started 

有一个人为创建的表空间,指向了其他位置: 
CREATE TABLESPACE tm_data 
  OWNER postgres 
  LOCATION '/home/pgsql_data'; 


目录拷贝过来之后 

修改postgres用户为简单密码tina 

测试之前那个insert能否成功: 
insert into t_black  (sample_hash, sample_crc32, sample_sha1, virusname, avl_m, software_english,file_size, conditions_type, 
sample_type, description,record_time, getname_time, developers, last_record_time, last_changename_time, source_id, ss_id, cps,level,keyhash, 
sensitive_strings, behavior_info, relation_info,version,ratio,"AVL_Embed", "AVL_Adware", program_name, versionname, record_time_int ) SELECT sample_hash, sample_crc32, sample_sha1, virusname, avl_m, software_english,file_size, conditions_type, 
sample_type, description,record_time, getname_time, developers, last_record_time, last_changename_time, source_id, ss_id, cps,level,keyhash, 
sensitive_strings, behavior_info, relation_info,version,ratio,"AVL_Embed", "AVL_Adware", program_name, versionname, record_time_int 
  FROM t_white where sample_hash ='1E0CB07CDC71B2F994F5D3EB51050E3A'; 
  
  
执行依然报错: 
ERROR:  invalid page header in block 27073 of relation base/21078/45300926 
********** 错误 ********** 

ERROR: invalid page header in block 27073 of relation base/21078/45300926 
SQL 状态: XX001 


再次像之前那样操作: 
dd if=/dev/zero of=/home/postgres/data/base/34780/42995 bs=8192 seek=1 count=1 conv=notrunc 

set zero_damaged_pages=on; 
vacuum full t_black20150301;  ---坏块就在这个表里面 

WARNING:  invalid page header in block 27069 of relation base/21078/45300926; zeroing out page 
WARNING:  invalid page header in block 27070 of relation base/21078/45300926; zeroing out page 
WARNING:  invalid page header in block 27071 of relation base/21078/45300926; zeroing out page 
WARNING:  invalid page header in block 27072 of relation base/21078/45300926; zeroing out page 
WARNING:  invalid page header in block 27073 of relation base/21078/45300926; zeroing out page 
查询成功但无结果,耗时: 43486 毫秒(ms)。 


tm_samples=# set zero_damaged_pages=on; 
SET 
tm_samples=# vacuum full t_black20150301;   ---原来vacuum full t_black无效,需要具体到某一个分区表。 
VACUUM 
tm_samples=# select count(*) from t_black; 
  count  
--------- 
3054944 
(1 row) 


再来执行insert试试: 
insert into t_black  (sample_hash, sample_crc32, sample_sha1, virusname, avl_m, software_english,file_size, conditions_type, 
sample_type, description,record_time, getname_time, developers, last_record_time, last_changename_time, source_id, ss_id, cps,level,keyhash, 
sensitive_strings, behavior_info, relation_info,version,ratio,"AVL_Embed", "AVL_Adware", program_name, versionname, record_time_int ) SELECT sample_hash, sample_crc32, sample_sha1, virusname, avl_m, software_english,file_size, conditions_type, 
sample_type, description,record_time, getname_time, developers, last_record_time, last_changename_time, source_id, ss_id, cps,level,keyhash, 
sensitive_strings, behavior_info, relation_info,version,ratio,"AVL_Embed", "AVL_Adware", program_name, versionname, record_time_int 
  FROM t_white where sample_hash ='1E0CB07CDC71B2F994F5D3EB51050E3A'; 
  
插入无报错,成功! 
  


下面再来试试剩下的3个表。 
tm_samples=# select count(*) from t_white;   --白表正常 
  count   
---------- 
25197129 
(1 row) 

tm_samples=# select count(*) from t_batch_sample; --这个表也正常,并都没有做vacuum操作。 
  count   
---------- 
30878635 
(1 row) 

tm_samples=# select count(*) from t_derivative; 
  count   
---------- 
70888497 
(1 row) 


----将有坏块的数据库迁移到磁盘正常的地方恢复,数据库只需要做一下vacuum操作,再次写入就没问题 




========================================================== 
补充: 
1、如何对一个磁盘进行格式化? 
mkfs命令 
使用方式 : mkfs [-V] [-t fstype] [fs-options] filesys [blocks] [-L Lable] 
  说明 : 建立 linux 档案系统在特定的 partition 上 
  参数 : 
  device : 预备检查的硬盘 partition,例如:/dev/sda1 
  -V : 详细显示模式 
  -t : 给定档案系统的型式,Linux 的预设值为 ext2 
  -c : 在制做档案系统前,检查该partition 是否有坏轨 
  -l bad_blocks_file : 将有坏轨的block资料加到 bad_blocks_file 里面 
  block : 给定 block 的大小 
  -L:建立lable 


[root@localhost beinan]# mkfs -t 文件系统  存储设备 
注: 
这里的文件系统是要指定的,比如 ext3 ;reiserfs ;ext2 ;fat32 ;msdos 等... ... 
设备比如是一个硬盘的分区,软盘,光驱等.. ... 在格式化分区之前,您得懂得如何查看硬盘分区情况,并有针对性的格式化; 
比如用 fdisk -l 来查看; 

您也可以把分区格式化成其它的文件系统;比如我们把 /dev/sda6格式化为ext3 、ext2、reiserfs、fat32、msdos 文件系统,命令格式如下; 
[root@localhost beinan]# mkfs -t ext3  /dev/sda6 
[root@localhost beinan]# mkfs -t ext2  /dev/sda6 
[root@localhost beinan]# mkfs -t reiserfs  /dev/sda6 
[root@localhost beinan]# mkfs -t fat32   /dev/sda6 
[root@localhost beinan]# mkfs -t msdos   /dev/sda6 


2、添加分区的自动挂载 
[root@pg home]# cat /etc/fstab 
UUID=f3b4f67a-7e8a-477f-8dca-a1d683aa8a57 /                       ext4    defaults        1 1 
UUID=9345bbc6-c984-4e42-9149-145649b6b753 swap                    swap    defaults        0 0 
UUID=e514fd36-4424-4984-952c-2661665f47c2 /home/pgsql                   ext4    defaults,noatime,nodiratime        0 0 
UUID=2bba011f-7bb2-43cd-ba83-6dcf0b459a33 /opt/db_backup                ext4    defaults,noatime,nodiratime        0 0 
tmpfs                   /dev/shm                tmpfs   defaults        0 0 
devpts                  /dev/pts                devpts  gid=5,mode=620  0 0 
sysfs                   /sys                    sysfs   defaults        0 0 
proc                    /proc                   proc    defaults        0 0 

UUID在哪里看呢? 
ls -l /dev/disk/by-uuid      #查看sdc1对应的uuid值 

使用df -T也可以查看到。 
[root@pg home]# df -T -h 
Filesystem     Type   Size  Used Avail Use% Mounted on 
/dev/sda2      ext4   104G   37G   62G  38% / 
tmpfs          tmpfs   16G     0   16G   0% /dev/shm 
/dev/sdc1      ext4   917G  529G  342G  61% /opt/db_backup 
/dev/sdb       ext4   939G  147G  745G  17% /home/pgsql 

cat /etc/fstab 
df -T -h 
mount 


3、有多余的表空间,且没有该表空间的目录依然可以正常启动数据库,如何将该表空间进行删除? 
could not open tablespace directory "pg_tblspc/30139248/PG_9.1_201105231": No such file or directory   经查看就是表空间tm_data 

postgres=# \db 
           List of tablespaces 
    Name    |  Owner   |     Location     
------------+----------+------------------ 
pg_default | postgres | 
pg_global  | postgres | 
tm_data    | postgres | /home/pgsql_data 
(3 rows) 

postgres=# drop tablespace tm_data; 
ERROR:  tablespace "tm_data" is not empty   --非空的表空间不能直接删除。 

select * from pg_tables where tablespace='tm_data';--查到有4张表在这个表空间下 
"public";"t_mobile_virus_url_event20150301";"postgres";"tm_data";f;f;f 
"public";"t_mobile_virus_url_event20150201";"postgres";"tm_data";f;f;f 
"public";"t_mobile_virus_url_event";"postgres";"tm_data";t;t;f 
"public";"t_mobile_virus_url_event20150401";"postgres";"tm_data";t;f;f 

先删除表,再删除表空间即可。 

4、如何查看数据库的ID和表的ID? 
select relfilenode from pg_class where relname='t_black';  --查看表的id 
select datid,datname from pg_stat_activity where datname='tm_samples';   --查看库的id     暂时没找到更直接的办法 
相关实践学习
使用PolarDB和ECS搭建门户网站
本场景主要介绍基于PolarDB和ECS实现搭建门户网站。
阿里云数据库产品家族及特性
阿里云智能数据库产品团队一直致力于不断健全产品体系,提升产品性能,打磨产品功能,从而帮助客户实现更加极致的弹性能力、具备更强的扩展能力、并利用云设施进一步降低企业成本。以云原生+分布式为核心技术抓手,打造以自研的在线事务型(OLTP)数据库Polar DB和在线分析型(OLAP)数据库Analytic DB为代表的新一代企业级云原生数据库产品体系, 结合NoSQL数据库、数据库生态工具、云原生智能化数据库管控平台,为阿里巴巴经济体以及各个行业的企业客户和开发者提供从公共云到混合云再到私有云的完整解决方案,提供基于云基础设施进行数据从处理、到存储、再到计算与分析的一体化解决方案。本节课带你了解阿里云数据库产品家族及特性。
相关文章
|
10天前
|
存储
tracker_query_storage fail, error no: 28, error info: No space left on device
tracker_query_storage fail, error no: 28, error info: No space left on device
7 0
|
8月前
|
Unix Linux AI芯片
4.3.3 info page
4.3.3 info page
50 0
|
索引
ES报错:“type“=>“cluster_block_exception“, “reason“=>“blocked by: [FORBIDDEN/12/index read-only / allow
ES报错:“type“=>“cluster_block_exception“, “reason“=>“blocked by: [FORBIDDEN/12/index read-only / allow
273 0
ES报错:“type“=>“cluster_block_exception“, “reason“=>“blocked by: [FORBIDDEN/12/index read-only / allow
|
程序员 算法框架/工具 Caffe
解决办法:error: 'size_t' does not name a type、unknown type name 'size_t'
解决办法:error: 'size_t' does not name a type、unknown type name 'size_t'
584 0
|
存储 关系型数据库 MySQL
浅析InnoDB Record Header及page overflow(1)
浅析InnoDB Record Header及page overflow
浅析InnoDB Record Header及page overflow(1)
|
存储 NoSQL 关系型数据库
浅析InnoDB Record Header及page overflow(2)
浅析InnoDB Record Header及page overflow
Whitelabel Error Page There was an unexpected error (type=Not Found, status=404). No message availab
Whitelabel Error Page There was an unexpected error (type=Not Found, status=404). No message availab
146 0